An innovative library for efficient LLM inference via low-bit quantization - View it on GitHub